Synaptic Plasticity, Memory and the Hippocampus

P e r s pec t i v e s

70. Vaynman, S., Ying, Z. & Gomez-Pinilla, F. Hippocampal 94. Rivera, S. M., Reiss, A. L., Eckert, M. A. & Menon, V. Acknowledgments BDNF mediates the efficacy of exercise on synaptic Developmental changes in mental arithmetic: evidence We would like to thank the National Institute on Aging (R01 plasticity and cognition. Eur. J. Neurosci. 20, for increased functional specialization in the left AG25,667, R01 AG25,032, R01 AG021,188) for their sup- 1030–1034 (2004). inferior parietal cortex. Cereb. Cortex. 15, port of our research and the preparation of this article. We 71. Ferris, L. T., Williams, J. S. & Shen, C. L. The effect of 1779–1790 (2005). would also like to thank A. R. Kramer for her help in crafting acute exercise on serum brain-derived neurotrophic 95. Coe, D. P., Pivarnik, J. M., Womack, C. J., Reeves, the article title. factor levels and cognitive function. Med. Sci. Sport M. J. & Malina, R. M. Effects of physical education Exerc. 39, 728–734 (2007). and activity levels on academic achievement in children. 72. Gold, S. M. et al. Basal serum levels and reactivity of Med. Sci. Sport Exerc. 38, 1515–1519 (2006). DATABASES nerve growth factor and brain-derived neurotrophic 96. Sallis, J. F. et al. Effects of health-related physical Entrez Gene: http://www.ncbi.nlm.nih.gov/entrez/query. factor to standardized acute exercise in multiple education on academic achievement: Project SPARK. fcgi?db=gene sclerosis and controls. J. Neuroimmunol. 138, Res. Q. Exerc. Sport. 70, 127–138 (1999). APOE | BDNF | IGF1 | TRKB| VEGF 99–105 (2003). 97. Martin, J. H. Neuroanatomy Text and Atlas. 2nd edn OMIM: http://www.ncbi.nlm.nih.gov/entrez/query. 73. Adlard, P. A., Perreau, V. M., Pop, V. & Cotman, C. W. (Appleton and Lange, Stanford Connecticut, 1996). fcgi?db=OMIM Voluntary exercise decreases amyloid load in a 98. Hall, C. D. Smith, A. L. & Keele, S. W. The impact of Alzheimer’s disease | Parkinson’s disease transgenic model of Alzheimer’s disease. J. Neurosci. aerobic activity on cognitive function in older adults: a 25, 4217–4221 (2005). new synthesis based on the concept of executive FURTHER INFORMATION 74. Cotman, C. W., Berchtold, N. C. & Christie, L.‑A. control. Eur. J. Cogn. Psychol. 13, 279–300 (2001). Charles H. Hillman’s homepage: http://www.kch.uiuc.edu/ Exercise builds brain health: key roles of growth factor 99. Bush, G., Luu, P. & Posner, M. I. Cognitive and labs/neurocognitive%2Dkinesiology/default.htm cascades and inflammation. Trends Neurosci. 30, emotional influences in anterior cingulate cortex. All links are active in the online pdf 464–472 (2007). Trends Cogn. Sci. 4, 215–222 (2000). 75. Prakash, R. et al. Cardiorespiratory fitness: a predictor of cortical plasticity in multiple sclerosis. Neuroimage 34, 1238–1244 (2007). 76. Berchtold, N. C., Chinn, G., Chou, M., Kesslak, J. P. & Cotman, C. W. Exercise primes a molecular memory for brain derived neurotrophic factor protein induction in the rate hippocampus. Neurosci. 133, 853–861 (2005). o p inion 77. Molteni, R. et al. Exercise reverses the harmful effects of consumption of a high-fat diet on synaptic and behavioral plasticity associated to the action of brain- derived neurotrophic factor. Neurosci. 123, 429–440 Synaptic plasticity, memory and (2004). 78. Stranahan, A. M. et al. Social isolation delays the positive effects of running on adult neurogenesis. the hippocampus: a neural network Nature Neurosci. 9, 526–533 (2006). 79. Barbour, K. A. & Blumenthal, J. A. Exercise training and depression in older adults. Neurobiol. Aging 26 (Suppl. 1), 119–123 (2005). approach to causality 80. Russo-Neustadt, A. A. & Chen, M. J. Brain-derived neurotrophic factor and antidepressant activity. Curr. Pharm. Des. 11, 1495–1510 (2005). Guilherme Neves, Sam F. Cooke* and Tim V. P. Bliss 81. Goldstein, D. B., Need, A. C., Singh, R. & Sisodiya, S. M. Potential genetic causes of heterogeneity of treatment effects. Am. J. Med.120 (Suppl. 1), S21–S25 (2007). Abstract | Two facts about the hippocampus have been common currency among 82. Etnier, J. et al. Cognitive performance in older women relative to ApoeE-epsilon4 genotype and aerobic neuroscientists for several decades. First, lesions of the hippocampus in humans fitness. Med. Sci. Sport Exerc. 39, 199–207 (2007). prevent the acquisition of new episodic memories; second, activity-dependent 83. Podewils, L. J. et al. Physical activity, APOE genotype, and dementia risk: findings from the cardiovascular health synaptic plasticity is a prominent feature of hippocampal synapses. Given this cognition study. Am. J. Epi. 161, 639–651 (2005). 84. Rovio, S. et al. Leisure time physical activity at midlife background, the hypothesis that hippocampus-dependent memory is mediated, and the risk of dementia and Alzheimer’s disease. Lancet Neurol. 4, 705–711 (2005). at least in part, by hippocampal synaptic plasticity has seemed as cogent in theory 85. Schuit, A. J. et al. Physical activity and cognitive as it has been difficult to prove in practice. Here we argue that the recent decline, the role of apoliprotein e4 allele. Med. Sci. Sports Exerc. 26, 772–777 (2001). development of transgenic molecular devices will encourage a shift from 86. Egan, M. F. et al. The BDNF val66met polymorphism affects activity dependent secretion of BDNF and mechanistic investigations of synaptic plasticity in single neurons towards an human memory and hippocampal function. Cell 112, 257–269 (2003). analysis of how networks of neurons encode and represent memory, and we 87. Kleim, J. A. et al. BDNF val66met polymorphism is suggest ways in which this might be achieved. In the process, the hypothesis that associated with modified experienced dependent plasticity in human motor cortex. Nature Neurosci. 9, synaptic plasticity is necessary and sufficient for information storage in the brain 735–737 (2006). 88. California Department of Education. California may finally be validated. physical fitness test: Report to the governor and legislature. Sacramento, California. Department of Education Standards and Assessment Division (2001). 89. Fields, T., Diego, M. & Sanders, C. E. Exercise is positively related to adolescents’ relationships and Multiple strands of evidence suggest an confirmed that the hippocampus is essential academics. Adolescence 36, 105–110 (2001). important role for the hippocampus in for the formation of new episodic memo- 90. Lindner, K. J. The physical activity participation- academic performance relationship revisited: episodic memory in animals and humans. ries and might also have a role in their perceived and actual performance and the effect of Most notable among human patients has long-term storage. Animal studies reveal banding (academic tracking). Ped. Exerc. Sci. 14, 155–169 (2002). been H.M., who as a young man suffered that controlled lesions, pharmacologi- 91. Maguire, E. A., Frith, C. D. & Morris, R. G. M. The from intractable epilepsy and underwent cal inactivation or molecular knockouts functional neuroanatomy of comprehension and memory: the importance of prior knowledge. Brain experimental surgery involving bilateral limited to the hippocampus result in either 122, 1839–1850 (1999). removal of the medial temporal lobe, a failure to learn or a loss of spatial mem- 92. Ansari, D. & Dhital, B. Age-related changes in the 2–5 6 activation of the intraparietal sulcus during including large parts of both hippocampi. ory . Electrophysiological recordings and nonsymbolic magnitude processing: an event-related The procedure left H.M. with an inability to molecular imaging studies in animals7,8, as functional magnetic resonance imaging study. J. Cogn. 9–11 Neuro. 18, 1820–1828 (2006). form new episodic memories (anterograde well as MRI imaging studies in humans , 93. Gobel, S. M., Johansen-Berg, H., Behrens, T. & amnesia), coupled with a substantial, but provide correlative evidence that episodic or Rushworth, M. F. Response‑selection‑related parietal activation during number comparison. J. Cogn. not total, loss of old memories (retrograde episodic-like learning and memory involves Neurosci. 16, 1536–1551 (2004). amnesia)1. Other cases since H.M. have hippocampal activity.

P e r s pec t i v e s

Polymodal sensory information

Hippocampus Entorhinal cortex

Temporoammonic Schaffer collaterals path Medial CA1 Lateral CA3 II III

V Dentate Associational/ gyrus commissural fibres Mossy fibres Perforant path

Modulatory input

Figure 1 | Basic anatomy of the hippocampus. The wiring diagram of the addition to the sequential trisynaptic circuit, there is also a dense associa- hippocampus is traditionally presented as a trisynaptic loop. The major tive network interconnecting CA3 cells on theNatur samee Re side.views CA3 | Neur pyramidaloscience input is carried by axons of the perforant path, which convey polymodal cells are also innervated by a direct input from layer II cells of the entorhinal sensory information from neurons in layer II of the entorhinal cortex to the cortex (not shown). The distal apical dendrites of CA1 pyramidal neurons dentate gyrus. Perforant path axons make excitatory synaptic contact with receive a direct input from layer III cells of the entorhinal cortex. There is the dendrites of granule cells: axons from the lateral and medial entorhinal also substantial modulatory input to hippocampal neurons. The three major cortices innervate the outer and middle third of the dendritic tree, respec- subfields have an elegant laminar organization in which the cell bodies are tively. Granule cells project, through their axons (the mossy fibres), to the tightly packed in an interlocking C‑shaped arrangement, with afferent fibres proximal apical dendrites of CA3 pyramidal cells which, in turn, project to terminating on selective regions of the dendritic tree. The hippocampus is ipsilateral CA1 pyramidal cells through Schaffer collaterals and to contra also home to a rich diversity of inhibitory neurons that are not shown in the lateral CA3 and CA1 pyramidal cells through commissural connections. In figure. For a full description of hippocampal anatomy, see REF. 90.

Synaptic plasticity in the hippocampus pharmacological agents to be rapidly ‘neurophysiological postulate’ proposes that The hippocampus has been a major washed on and washed off and allowing connections between co-active neurons experimental system for studies of synaptic intracellular and patch-clamp recordings. are strengthened through mechanisms plasticity in the context of putative informa- In addition, hippocampal neurons can be of synaptic plasticity, so that subsequent tion-storage mechanisms in the brain. Its cultured28,29, either as transverse ‘organo- activation by incoming stimulation of only simple laminar pattern of neurons and typic’ slices or as populations of dissociated a sub-component of the assembly will lead neural pathways (FIG. 1) enables the use of neurons, for periods of months, facilitating to activation of the whole assembly, thereby extracellular recording techniques to record molecular manipulations such as over recapitulating the activity elicited by the synaptic events for virtually unlimited peri- expression or RNAi-based knock-down of original event. (LTP is a Hebbian process, ods in vivo12. The much-studied model of specific proteins. These in vitro techniques since its induction requires coincident synaptic plasticity, long-term potentiation13,14 have greatly enhanced our understanding activity of the pre- and postsynaptic neu- (LTP; see FIG. 2a), was first identified in of the molecular mechanisms that underlie rons.) The immediate problem is to identify the hippocampus and has been extensively synaptic plasticity15,30. In the hippocampus it such cell assemblies in the hippocampal characterized using electrophysiological, has been possible to track effects such as the encoding of memory. biochemical and molecular techniques15. phosphorylation of a protein at a specific Several recent studies have detected LTP- residue at multiple levels of organization, Place cells like synaptic changes in the hippocam- from isolated synaptic membranes all the Single-unit recordings from neurons in pus16,17 (FIG. 2b) and the amygdala18 following way through to the behavioural analysis the hippocampus of freely moving rodents learning. Other forms of activity-dependent of intact animals with specific molecular reveal that pyramidal and granule cells show plasticity have been found, including defects31. Nevertheless, the larger picture a preference for firing in a particular loca- long-term depression (LTD)19, EPSP-spike of how synaptic plasticity in extensive tion of an explored environment, regardless (E‑S) potentiation20,21, spike-timing-dependent networks of cells leads to the storage and of the direction from which the animal plasticity (STDP)22, depotentiation23–25 and recall of information remains dimly illumi- enters the location33 (BOX 1). Hundreds de‑depression25,26. The transverse hippo nated. The Canadian psychologist Donald of such ‘place cells’ fire in concert as a rat campal slice preparation27 (FIG. 2a) has been Hebb posited a role for such assemblies as reaches a particular location, and place of major importance to this field, enabling engrams or memory traces32. His famous cells fire in sequence as the animal moves

P e r s pec t i v e s

through a series of locations in a given a Tetanized Rec Untetanized b 34,35 environment , suggesting that a network input input Inhibitory avoidance (IA) of pyramidal cells can also serve as a cell assembly to encode and store a neural Sch CA1 representation of space. Mossy fibres LTP and learning: approaches to causality DG pp The SPM hypothesis. The presumptive CA3 (from EC) causal link between synaptic plasticity and memory has been formalized by Morris and colleagues as the synaptic plasticity and Pre- Post- Pre- and IA-induced LTP Tet-induced LTP tetanus tetanus post-tetanus 1 + 2 2 + 3 memory (SPM) hypothesis:

Activity-dependent synaptic plasticity Tetanized Unchanged is induced at appropriate synapses input input during memory formation, and is both necessary and sufficient for the information storage underlying the type Untetanized Enhanced input 1 mV input 1 mV of memory mediated by the brain area 5 ms 5 ms in which that plasticity is observed36.

75 It is now over 30 years since the first 200 Re-normalize description of LTP in the hippocampus, 150 50 3 20 years since the first attempt to use phar- 100 25 2 macological tools to dissect the relation- 1 ship between LTP and memory, and over 50 sponse (% change) (% sponse change) (% sponse 0

10 years since the first knockout studies re re 0 –25 were published. Even though the SPM IA Tet Tet Tet hypothesis, or a similar model, is enshrined –50 –50 Synaptic –1 0 1 2 3 4 5 6 7 8 Synaptic 0 1 2 3 4 in most neuroscience textbooks, the issue Time (hours) Time (hours) is far from resolved. We next consider the reasons for this impasse, and ask what new Figure 2 | Long-term potentiation (LTP) in vitro and in vivo. a | Extracellular recordings of LTP induced by tetanic stimulation of the Schaffer-commissural projection (Sch) to CA1 pyramidal cells approaches are needed if the relationship is Nature Reviews | Neuroscience ever to be unravelled. in a transverse hippocampal slice (shown as a schematic in the top panel). Hippocampal slices can be kept healthy for many hours if a steady flow of oxygen and artificial cerebrospinal fluid is supplied. The laminated organization of the hippocampus lends itself perfectly to extracellular recording Testing necessity. In order to establish the techniques, allowing selective pathways to be stimulated and the evoked synaptic responses gener- necessity of synaptic plasticity (taking LTP ated by a population of target neurons to be monitored for prolonged periods of time. The middle as our exemplar) for information storage, panel shows typical synaptic responses recorded from the apical dendritic region of the CA1 subfield the ideal experiment would be an interven- following stimulation of the Schaffer-commissural pathway. Two metal stimulating electrodes are tion that completely blocked the induction placed on either side of the recording electrode to evoke responses in overlapping populations of or expression of LTP in the hippocampus pyramidal cells through different sets of synapses. A tetanus (a brief, high-frequency train of electri- while doing nothing else. The twin prob- cal stimuli) can be used to induce LTP lasting for many hours in the tetanized pathway (bottom panel, lems in any real-life experiment lie in the closed circles); the second, control pathway (open circles) receives only test stimulation and is not precise spatial targeting of the blockade potentiated following the tetanus to the experimental pathway. This demonstrates an important property of LTP, namely input specificity. b | In vivo LTP induction by learning17. Synaptic responses and in the need to affect ‘nothing else’. At from multiple locations can be recorded in area CA1 of freely moving animals using an array of record- first glance, the early observation that infu- ing electrodes and a single stimulating electrode (examples in middle panel). Rats were trained in an sion into the hippocampus of the selective inhibitory avoidance (IA) task, a hippocampus-dependent form of single-trial learning in which a NMDA (N-methyl-d-aspartate)-receptor rodent avoids entering a dark arena where it has received a footshock (top panel). IA training leads blocker APV (2-amino-5-phosphonovaleric to a rapid increase, lasting for hours, in the amplitude of evoked responses in some of the recorded acid) profoundly impairs learning and pathways (green circles in lower panel) but not in others (red circles). Training-dependent synaptic recall in the Morris water maze4 is a compel- enhancement (bottom panel, arrow IA) occludes LTP induced by delivering tetanic stimulation (bot- ling validation of the hypothesis: the drug tom panel): compare the degree of potentiation induced by tetanic stimulation (arrow Tet) in the is applied directly into the hippocampus pathways that were enhanced by training (green circles) to the pathways that were unchanged (red and blocks LTP without affecting basal circles). The numbers 1, 2 and 3 indicate the times at which sample responses were obtained from inputs that were either enhanced (green) or unchanged (red) following learning. Note that post-IA synaptic transmission. In a crucial recent responses are re-normalized before tetanus-induced LTP. Superimposed responses in the middle experiment, it was shown that inhibition of panel show effects of learning (1+2) and the subsequent effects of delivering three episodes of tetanic the active form of the protein kinase PKMζ stimulation (2+3). These results suggest that experience-dependent synaptic enhancement uses the by infusion into the hippocampus of its same molecular mechanisms of expression as tetanus-induced LTP. DG, dentate gyrus; EC, entorhinal specific inhibitor, ZIP (myristoylated zeta- cortex; pp, perferant path. Part a modified, with permission, from REF. 91  (2003) Blackwell Science. pseudosubstrate inhibitory peptide), can Part b reproduced, with permission, from REF. 17  (2006) American Association for the Advancement impair spatial memory and block LTP, even of Science.

P e r s pec t i v e s

cell processes other than LTP; even induc- Box 1 | Plasticity in place cells ible systems require several days to take Place cells are hippocampal pyramidal or Control box Training box effect, during which time compensatory granule cells that fire action potentials in mechanisms can develop. The most com- particular locations (place fields) in an pelling transgenic experiment so far is the environment and which thus collectively carry reversible inactivation of the NMDA recep- information about the animal’s moment-to- tor subunit NR1 in the CA1 subfield of the moment position. hippocampus37. Here the gene product is Ensembles of place cells probably serve as our best working model of hippocampal function. directly responsible for the induction of However, they do not observe any obvious LTP, and the molecular deficit is precisely spatial topography and certainly do not defined, inducible and reversible. It also conform to a two-dimensional topographical seems to be mostly confined to pyramidal map76,77. Adjacent place cells in the cells in area CA1. In this mouse both LTP hippocampus can encode locations separated and spatial learning are suppressed, imply- by great distances in an environment and, also, Before ing that the presence of NR1 receptors in an individual place cell’s receptive field can be area CA1 is necessary for spatial learning very different from environment to to occur. Is it equally safe to conclude that environment. This observation is consistent LTP in area CA1 is necessary for spatial with the idea that spatial memory is encoded in a distributed fashion in the hippocampus78. Fear learning? The answer is no, as the blockade conditioning Plasticity of place cells has been observed as a of the NMDA receptor is known to affect remapping of either their firing rates or their several other processes, including the receptive fields when cues in an environment79, induction of E-S potentiation38 and certain or the shape of an environment, are forms of LTD19, and to reduce postsynaptic changed80,81. Remapping can also be triggered responses during short bursts of high- After by a discrete learning event in the same, frequency activation39; any or all of these 82 unchanged environment — in a form of processes might contribute to information Pavlovian conditioning called contextual fear storage. The closer the experimental conditioning (middle and lower pairs of panels). intervention gets to LTP itself, however, Here, an electrical footshock is applied as an unconditional stimulus while the environment effectively acts as a conditional stimulus that, after the more confident we can be that there Nature Reviews | Neuroscience training, can itself elicit a behavioural freezing response. The middle and lower pairs of panels in is a causal link between LTP-like synaptic the figure show the firing rates of one place cell in two different environments. The cell’s place field plasticity and learning and memory. was stable in a control environment (left-hand panels) but remapped from the north east to the Nevertheless we have to conclude that, south west of the experimental chamber (right-hand panels) after contextual fear conditioning. despite the wealth of experimental support, Firing rates in the environment are colour coded (redder colours indicate higher firing rates). definitive evidence that LTP is necessary Assuming that this form of remapping depends on hippocampal plasticity, it could serve as an for hippocampus-dependent learning is intermediate electrophysiological assay for effective silencing, erasure or re-installation of memory still lacking. in the proposed experiments illustrated in FIGS 3,4. Evidence that remapping requires hippocampal plasticity has come from analysis of subfield-specific knockouts. Place cell activity is disrupted in Learning without hippocampal LTP? animals with CA1-specific knockout of the NMDA receptor subunit NR1, such that receptive fields do not retain strong location specificity and ensembles of cells with similar receptive fields are not Given that a single negative result could correlated in their firing, consistent with the disruption of a functional representation of space83. ostensibly disprove the necessity arm of Similar results were obtained with perfusion of an NMDA-receptor antagonist into the the SPM hypothesis, a potentially more hippocampus84. In CA3-specific knockouts of NR1, CA1 place cells have normal place fields in powerful result would be one in which LTP familiar environments but enlarged, unrefined place fields in novel environments85, suggesting a is suppressed yet learning is unaffected; role for plasticity at CA3 recurrent collateral synapses in remapping of place fields. Place cell in such a case, whether or not other pro remapping in area CA3 is also disrupted when NR1 expression is deleted in the dentate gyrus54. cesses are affected, the conclusion can be Figure reproduced, with permission, from REF. 82  (2004) Society for Neuroscience. drawn that LTP is not necessary for that particular form of learning. One example is the ‘upstairs/downstairs’ water maze when the inhibitor is administered days 4-isoxazole propionic acid) receptors experiment40, in which rats were trained in after the acquisition of the memory or the (a candidate mechanism for the expression one maze (on a lower floor of the labora- induction of LTP, again without affecting of LTP30) might be necessary for the full tory building) and subsequently were able baseline synaptic transmission5. However, expression of amygdala-dependent cued fear to learn and retain information about the in both cases it is impossible to be certain conditioning18. Again, it is difficult to exclude location of the hidden platform in a second that the drug has not spread outside the the possibility that effects on processes upstairs maze, even when infused with hippocampus and is not having some effect unrelated to the maintenance of LTP cause the NMDA-receptor antagonist APV. This other than blocking the induction or main- the learning impairment. experiment is important because it sug- tenance of LTP4 (see below). The use of The situation is not obviously improved gests that, at least in some circumstances, viral vectors to interfere with the process of in most experiments using genetically conventional NMDA-receptor-dependent glutamate-receptor trafficking suggests that engineered mice, as the effects of knock- LTP is not required for the acquisition and membrane insertion of GluR1‑containing ing out a transcription factor or a protein storage of hippocampus-dependent AMPA (α-amino-3-hydroxy-5-methyl- kinase, for instance, will certainly affect reference memory.

P e r s pec t i v e s

The GluR1 knockout. Another potential that implicit motor learning is mediated by of NR1 in granule cells of the dentate gyrus counter-example to the predictions of the synaptic plasticity in the cerebellar cortex impairs both working memory in a radial arm SPM hypothesis is the GluR1-knockout and/or the deep cerebellar nuclei46–49. These maze55 and discrimination of context in fear mouse, which showed a total absence of structures are organized as two-dimensional conditioning54, suggesting that a failure of conventional tetanus-induced LTP in area topographical maps of the body, and it pattern separation occurs. CA1 without any impairment in acquisition is possible to target specific microzones These results are consistent with the or recall in the standard, reference memory that mediate particular skeletal muscular much-cited model of subregional process- version of the Morris water maze44. A responses50,51. The best-studied example of ing in hippocampal function developed by problem with reaching a firm conclusion this functional organization is probably David Marr56. The situation is unresolved from such an apparently definitive result is classical conditioning of the nictitating in area CA1 where, as we have seen, the evi- that it is very hard to be sure that LTP has membrane/eyeblink response in rabbits49,52. dence does not allow us to decide whether or in fact been abolished. The experimenter With such regular, tractably organized and not LTP is necessary for reference learning has available a range of protocols to induce well-characterized circuitry, there might in the water maze. The role of non-NMDA- LTP, but we do not know what protocols the be some hope of developing interventional receptor-dependent plasticity at mossy fibres hippocampus itself is using. It might be that assays to test the causal role of synaptic also remains to be clarified. the behaving animal was able to generate plasticity in motor learning, for example, by ‘sharp-wave ripples’, a naturally occurring selectively erasing or installing conditioned Network approaches high-frequency waveform generated by eyeblinks. By contrast, the functional A direct test of the proposition that the neu- synchronous firing of CA3 pyramidal cells organization of the hippocampus is less ral representation of a memory is encoded in that can facilitate the induction of LTP41. understood, largely owing to a lack of two- the network of neurons containing synapses The initial GluR1-knockout paper reported dimensional topography (BOX 1). However, that were modified during the acquisition of that LTP induced by a brief burst of 100 Hz it might be possible to address the SPM that memory would be to ask whether the stimulation was absent in the Schaffer- hypothesis in the hippocampus by adopt- memory is lost when only these neurons are commissural pathway, but only reduced to ing a set of experimental strategies that are selectively inactivated. about 50% of controls in the perforant path42 largely blind to functional organization. We (FIG. 1). Subsequent analysis revealed that, outline some of our suggested approaches Exploiting immediate-early genes. A feasible even in the Schaffer-commissural pathway, below but, before doing so, we need to sum- approach to silencing potentiated neurons LTP could be induced using a theta-burst marize what has been learned from mutant at the network level would be to design a pairing protocol, in which presynaptic mice engineered to express region-specific transgenic mouse in which a promoter for stimulation at 5 Hz was paired with synchro- and/or inducible transgenes. a plasticity-marker gene drives the expres- nous depolarization of the CA1 pyramidal sion of a protein that reduces the excitability cell43. (Theta-burst stimulation mimics the Subregion-specific deletion of NR1 of the cell. No such plasticity-marker genes frequency of theta waves that are generated Over the past decade several studies have have been identified that are expressed when in the hippocampus of rodents as they used the enzyme Cre recombinase, driven and only when synaptic plasticity is induced. explore an environment44,45.) by subregion-specific promoters, to restrict However, the expression of several immedi- In summary, the GluR1 animal has not deletion of the gene that encodes the ate-early genes (IEGs), including those that disproved necessity, and it is hard to see how NMDA-receptor-subunit NR1 to particular encode c‑Fos, Zif268, Arc/Arg3.1 and Homer, any other transgenic mouse would get around subfields of the hippocampus, giving are strongly upregulated by LTP-inducing this problem. Finally, we note that even if LTP rise to different cognitive impairments. protocols both in vitro and in vivo, as well as is not necessary for learning, it might never- Although, as we have seen, NR1 receptors by behavioural training (FIG. 3a) (reviewed in theless be the brain’s default choice when it is in pyramidal cells of area CA1 seem to be Ref. 57). Both Arc/Arg3.1 (Refs 58,59) and available, as in the normal brain. essential for normal performance on the the transcription factor Zif268 (Ref. 60) are reference memory version of the Morris required for the maintenance of LTP that Testing sufficiency. Turning to sufficiency, water-maze2,37, animals can perform these lasts for several days and for the stability of can we devise an experiment in which a tasks as well as control littermates if the NR1 long-term memories. A knock-in mouse novel memory is installed by inducing LTP deletion is confined to pyramidal cells in line has been generated in which the coding at a selection of hippocampal synapses? area CA3 (Ref. 53) or granule cells in the den- sequence of the Arc/Arg3.1 gene was replaced The answer is that we know so little about tate gyrus54,55. We can therefore be confident with that of the gene that encodes green fluo- how episodic memories are encoded in the that NMDA-receptor-mediated LTP in the rescent protein (GFP). GFP expression in the hippocampus or in the neocortex that even dentate gyrus54,55 and in area CA3 (Ref. 53) is visual cortex of these mice could be tracked if we had the experimental tools to modulate not necessary for the acquisition and storage in vivo, and was regulated by light exposure synaptic weights at a spatially distributed set of reference spatial memory, as this form of in an NMDA-receptor-dependent manner of hippocampal synapses, we would have no LTP is impossible in regions where NR1 (FIG. 3b), faithfully mimicking the expression idea how to go about selecting which syn- has been deleted and yet these animals can of Arc/Arg3.1 (Refs 59,61). In a recent study62, apses to modify. Conceivably, the situation learn. However, this is not to say that these the promoter of the gene that encodes c-Fos is more tractable for other forms of memory subregions have no role in spatial memory. was used to generate a mouse line that per- in the brain. In the cerebellum, for instance, Animals with deletion of NR1 in area CA3 mits long-lasting genetic tagging of activated there is a regularly organized circuit that have subtle defects in pattern completion, such neurons. This study suggested that neurons delivers relatively unprocessed somatosen- that they are unable to use partial presenta- that are activated in the amygdala during sory and motor information to the Purkinje tion of external cues to recall the position contextual fear conditioning are re-activated cells of the cerebellar cortex. It is believed of a hidden platform53. By contrast, deletion during retrieval of the memory.

P e r s pec t i v e s

a b These and other studies63,64 show that it e is possible to activate transgenes specifically enc in circuits that are activated by particular esc behaviours. It follows that we should be able to design molecular devices that reversibly activate or inactivate hippocampal neuronal

ect GFP fluor spiking, using the promoters of IEGs to

Dir drive the expression of membrane proteins Control Novel Novel Novel Novel exploration chamber chamber chamber that generate appropriate changes in excit- (dark) (light) (MK801 + light) ability. Using these devices, it should be c AlstR-expressing cells + 10 nM Alst possible to test the causal role of plasticity in distributed networks of neurons in the hippocampus during learning and memory. 0 This strategy takes us closer to establishing

Vm (mV) the importance of Hebbian cell assemblies in information storage. By exploiting the –50 transient and activity-dependent expression of IEGs, it should be possible to dis- 50 pA 1 s criminate between cell assemblies: the IEG promoter will drive gene expression only in d Memory cell expressing AlstR those cell assemblies that represent a newly Arc/Arg3.1 pro AlstR EGFP acquired memory, leaving other hippocampal Memory cell no longer expressing AlstR neurons unaffected (FIG. 3d).

Silenced memory cell Reversible activation or inactivation Task A Task B Designs for a new generation of engineered ligand- and light-gated membrane receptors, which can be used to initiate or inhibit neural activity, have come to the fore in the past year or two.

Ligand-gated systems. Many invertebrates express a gated Cl– channel that is kept open by ivermectin (IVM), a drug that is widely used for the control of worm infections. Day 1 Day 7 Day 7 + Alst In the presence of low concentrations of Figure 3 | A strategy for silencing a hippocampal cell assembly encoding a particular ivermectin, cultured hippocampal neurons memory. The approach depends on genetic constructs in which promotersNatur frome Revie activity-dependentws | Neuroscience expressing this channel are held near the genes (such as those that encode Arc/Arg3.1 or Zif268) are used to drive the expression of trans- hyperpolarizing Cl– reversal potential, and genes specifically in recently potentiated cells. These transgenes can be used to silence activated thus are effectively silenced65. The channel cell assemblies. a | Arc/Arg3.1 (immunostained in red) is activated in a subset of hippocampal neu- is composed of two subunits, α and β, both rons (in this figure, CA1 pyramidal cells) when an animal explores a novel environment92 (scale bar of which must be expressed in the same 100 µm). b | Green fluroescent protein (GFP) mirrors the endogenous expression of Arc/Arg3.1 in a cell to generate an IVM‑sensitive current65. genetically engineered mouse in which the expression of GFP is controlled by the Arc/Arg3.1 gene Inspired by the potential of this approach, we promoter59. In this example, expression in the primary visual cortex is upregulated by light exposure. attempted to generate a transgenic mouse that The NMDA (N-methyl-d-aspartate) receptor antagonist MK801 blocks this effect (scale bar 40 µm). c | The left-hand panel shows a confocal micrograph of a section obtained from the spinal cord of a would express the IVM-binding α‑subunit transgenic mouse in which the engrailed gene promoter had been used to drive expression of the under the control of the Arc/Arg3.1 pro- allatostatin receptor (AlstR) in a specific subtype of interneuron (labelled in green, reflecting expres- moter (conferring activity-dependence on sion of GFP). The right-hand panel shows a current-clamp recording from a labelled interneuron: in the transgene) and the β-subunit under the the presence of 10 nM allatostatin (Alst), neurons have a higher threshold for triggering action control of the a‑calcium/calmodulin- potentials and therefore are effectively inactivated68. d | A way in which to abolish specific memories dependent protein kinase II promoter (to while sparing others. The activity-dependent Arc/Arg3.1 gene promoter is used to drive the restrict expression to the postnatal forebrain). expression of the allatostatin receptor. Cells that, as a result of training in Task A, acquire potenti- Further spatial selectivity would be achieved ated synapses (green cells in left panel) will express the receptor and can potentially be silenced by the direct injection of IVM into the by perfusion with allatostatin. Silencing is dependent on allatostatin, but also on the presence of the allatostatin receptor on the cell surface. After a certain period of time the receptors will be hippocampus. Our attempts have so far been internalized and degraded (red cells, day 7). Recent memory, activating a different, possibly overlap- jeopardized by inadequate expression levels. ping, population of cells (Task B, green cells) should therefore be abolished when these cells become However, this method has recently been suc- silent (black) in the presence of allatostatin, whereas remote memories (red cells) should be spared. cessfully used to silence cells in the striatum Part a reproduced, with permission, from REF. 92  (2005) Society for Neuroscience. Part b repro- by viral infection66, although the silencing duced, with permission, from REF. 59  (2006) Elsevier Science. Part c reproduced, with permission, that was achieved had a slow activation and from REF. 68  (2006) Macmillan Publishers Ltd; courtesy of M. Goulding. inactivation time scale. The approach may

P e r s pec t i v e s

therefore be more suitable for use in the con- Box 2 | The hippocampus and global versus local approaches to neural silencing text of neural development or pathological processes. Memory has many forms and is distributed across many brain regions. Although the hippocampus is required for the formation of episodic or episodic-like memory, it remains unclear whether A more promising approach might be the hippocampus itself acts as a memory store (and, if so, for how long). In some descriptions the to exploit metabotropic signal transduc- hippocampus, although a necessary component of the memory system, does not itself store tion pathways to amplify the effect of memories. Rather, it acts as an indexing device in which the hippocampal cell assembly that is the transgene on cell excitability. When activated during learning has access, through reciprocal cortical–hippocampal connections, to transfected into mouse neurons, a the neocortical neural networks where episodic memories are actually stored86. How long Drosophila melanogaster G‑protein-coupled memories are stored or indexed in the hippocampus is also a topic of debate. In many cases the receptor for the peptide hormone allato- presence of the hippocampus seems only to be required during an initial period, lasting a few statin (AlstR) can couple to mammalian weeks in rodents and perhaps a few years in humans, during which the permanent memory is G‑protein-activated inwardly rectifying K+ gradually ‘consolidated’ in the neocortex. The patient H.M. provides one example of such a case. In channels (GIRK or Kir3 channels)67. This other circumstances however, the presence of the hippocampus is required for extended periods, and perhaps indefinitely, as seems to be the case with reference spatial memory in rodents. For a allows rapid, reversible hyperpolarization fuller discussion of these issues, see REFS 87–89. and hence silencing of allatostatin‑receptor- Whatever the exact distribution of the engram or memory store for a given type of memory, our expressing neurons in response to allato proposed strategy involving the exploitation of immediate-early genes (IEGs; see main text) is statin. For example, when transgenic mice sufficiently flexible to encompass these multiple possibilities. The approach would allow us to expressing the allatostatin receptor in a operate at a systems level, by delivering an exogenous trigger (ideally a systemically deliverable specific class of spinal cord interneurons were ligand, such as ivermectin, that can cross the blood-brain barrier). In this sense the experiment can treated with allatostatin, the neurons be done ‘blind’ — without the whereabouts of the encoding networks in the brain being known — were quickly and reversibly inactivated68 because the procedure will inactivate all potential components of a memory. However, the (FIG. 3c). Applied to the silencing of cell strategy is not limited to operating blind at a systems level. If the IEG exploitation experiments we assemblies in the hippocampus, this propose were conducted in a targeted fashion using local hippocampal infusions of the silencing ligand ivermectin, rather than systemic infusion, we would hope to parse out the hippocampal approach would require a transgenic animal contribution to episodic-like memory. A similar approach could be used throughout the nervous in which expression of the allatostatin system to define causal roles for individual structures in different forms of information storage. In receptor is driven by an activity-dependent this way, the problem of whether the hippocampus stores memories or merely indexes them also promoter (FIG. 3d). Spatial specificity would becomes addressable, by monitoring the behavioural effects of successively silencing alternative be gained by infusing allatostatin to the storage sites. hippocampal region of interest (BOX 2). At least two other methods for ligand-controlled inactivation of neurons have been success- would drive the expression of light-dependent Is synaptic plasticity sufficient for memory? fully used in behaving mice69,70, and all signs ion channels, has been proposed for the Can synaptic plasticity alone be used to are that these methods will continue to study of neural networks that are implicated build a memory? To answer this question we develop at a rapid pace. in disease models72. need to be able to mimic the natural process of synaptic plasticity by artificial means, at Light-gated systems. A whirlwind of Memory erasure. Another approach to only the subset of synapses that would be interest has been generated recently by testing the necessity of LTP for learning involved in storing a particular memory. strategies that rely on light exposure to and memory would be to attempt to erase Our aim is to make memory without the modulate neural activity. Chlamydomonas memory by selectively reversing experience- need for learning (memory mimicry). reinhardtii channelrhodopsin 2 (ChR2) dependent plasticity. As discussed earlier, In the current state of knowledge this is is a cation channel that opens on perfusion of the specific PKMζ inhibitor, not feasible, and it is unlikely to become so exposure to blue light (thereby depolar- ZIP, erases hippocampus-dependent mem- any time soon. However, one can imagine izing neurons that express it), whereas ory. ZIP only targets activated synapses, but ways in which it might be possible to recreate a Natronomonas pharaonis halorhodopsin it does not appear to differentiate between lost memory. Two such thought‑experiments (NpHR) is a Cl– pump that is activated by recent and old memories. One conceptually are described in FIG. 4. In both cases a hippo yellow light (and which thus can hyper simple approach to reversing change at campus-dependent memory is formed by a polarize neurons in which it is expressed)71. recently potentiated synapses would be to standard training procedure. The memory is The use of light as a switch allows precise use the phenomenon of depotentiation. In then erased, but subsequently re-installed by temporal control but presents challenges in area CA1, synapses that have recently been exploiting the knowledge gained about the getting enough light through the skull and potentiated can be depotentiated by low fre- synaptic changes that occurred during the into the brain area expressing the light‑ quency stimulation24. Depotentiation occurs original learning episode. sensitive ion channels. However, major only at those synapses that were potentiated strides have been made in solving this in the preceding few minutes, after which The very large multi-electrode array. The problem72,73. In one particularly impres- LTP becomes stabilized and resistant to first experiment relies on gaining access sive study, the selective optical activation depotentiation. Thus, depotentiating stimu- to a large enough assembly of pre- and of hypocretin-producing neurons in the lation could, in principle, be used to erase postsynaptic neurons to monitor plastic- hypothalamus, an area that lies deep inside LTP specifically in the cell assemblies that ity between pairs of cells participating in the brain, was shown to awaken sleeping represent a recently acquired memory. As we encoding a new memory, each with one animals73. Again, in our context, ChR2 or discuss below, memory erasure might also partner in area CA3 and one partner in NpHR would be driven by IEG promoters. prove useful in testing the sufficiency arm of area CA1 (FIG. 4a). Initial cross-correlation The same strategy, in which an IEG promoter the SPM hypothesis. of action potentials from each CA3–CA1

P e r s pec t i v e s

pair before learning will detect rare pairs of in which LTP or LTD is induced by appro- The ideal transgenic mouse. The second cells that are connected, and the amplitude priately ordering the sequence of pre- and experiment (FIG. 4c) makes use of an imagi- of the cross-correlogram will be a measure of postsynaptic spiking, could then be applied nary but not wholly implausible molecular the strength of the synapse that links the two to re-tune synapses back to their memory device that serves as a free-standing cells (FIG. 4b); comparison of the peak before state. In this way, we could selectively reverse LTP device. This alien device might be, for and after learning will indicate which of the synaptic changes that underlie the storage example, a Ca2+ channel that when operated connected pairs are part of the cell assembly of one memory without affecting either by an exogenous ligand allows the permea- that encodes the memory. Depotentiation basal transmission or synaptic changes that tion of sufficient Ca2+ to trigger the native by immediate post-training application of subserve the storage of other memories, and LTP induction mechanisms (FIG. 4d). The low-frequency stimulation would erase the then, subsequently and on demand, reinstall strategy is to target the free-standing LTP memory. Spike‑timing-dependent plasticity, the lost memory. device to recently potentiated synapses

a b Before After After After CA3 CA1 learning learning depotentiation repotentiation

A A 1 1 0 τ (ms) B 2 B 3 3 C C 2

c d Alien ligand e

Forgetful mouse ×

Free-standing LTP device Ca2+

Arc/Arg3.1 pro Free-standing LTP device LTP Learning-associated synapses only

Figure 4 | Thought experiments: erasing and re-installing a hippocam- same aim. With currently available technology, the best way to gain access to Nature Reviews | Neuroscience pus-dependent memory. a | The first thought-experiment requires a very potentiated synapses is by first training the animal to form a memory that is large array of metal electrodes to monitor the spike activity of each cell in transient. One way to achieve this is to use mutant animals that fail to form areas CA3 and CA1 (connectivity is unidirectional, from CA3 to CA1); the long-term memory, termed here ‘forgetful mice’, such as mice in which same electrodes can be used to stimulate each cell individually. Connectivity Arc/Arg 3.1 (Ref. 58), Zif268 (Ref. 60) or α/δCREB93 have been knocked out. An is sparse, and the great majority of CA3–CA1 cell pairs are not connected. immediate-early gene promoter can be used to drive transcription of a b | Cross-correlation of spontaneous activity will identify connected pairs of molecular LTP device in recently activated synapses (shown in red in e). The CA3 and CA1 cells (here, A to 1, B to 3 and C to 2). The cross-correlogram plots transcript could encode, for instance, an exogenous ligand-gated Ca2+ chan- the number of spikes emitted by a given CA1 cell during a given time interval nel (d). Infusion of the exogenous ligand would activate the Ca2+ channel (τ, τ + δτ) after each action potential of a given CA3 cell; a peak at a delay τ of (free-standing LTP device) in only those synapses that had recently been a few milliseconds suggests that the two cells are monosynaptically con- potentiated, inducing further potentiation in those synapses and thus re- nected. Following learning, the SPM hypothesis predicts that a subset of installation of a memory in an animal in which memories were normally only synapses will be potentiated (some perhaps will be depressed); these pairs will transient. e | An important development that existing technologies do not yet be identified by an increase (or decrease) in the peak of the cross-correlogram. allow is the targeting of transgenes to the specific synaptic sites that have Each of these affected synapses can be either depotentiated by low-frequency undergone plasticity. Arc mRNA94 and protein95 are selectively transported to stimulation (which has no effect on unpotentiated synapses) or re-potentiated dendritic regions containing recently potentiated synapses, and possibly to by appropriately timed spike-timing-dependent potentiation. Returning all the potentiated synapses themselves. The molecular mechanics behind the synaptic strengths to baseline by depotentiation should abolish the memory. putative ‘tagging’ of synapses that allows them to capture recently synthe- At an arbitrary later time, the memory can be reinstalled by re-potentiating or sized proteins remains elusive. When we learn how synapses do this, we may re-depressing the affected synapses by appropriately timed spike-timing- be in a position to target exogenous proteins, including free-standing LTP dependent plasticity. c | An attempt to use molecular genetics to achieve the devices, specifically to synapses that are activated during learning.

P e r s pec t i v e s

expressing synaptic tags, as postulated by of neurons that are activated during learn- memory, allowing questions of causality Frey and Morris74. The LTP device would ing. We predict that new technologies will to be addressed at the level of what Hebb be driven by a promoter from an IEG, such allow the silencing of the subset of hippo called the cell assembly. Circuit-specific as from the gene that encodes Arc/Arg3.1, campal neurons that encode a particular memory erasure would demonstrate that and the transcript would have to contain at least a dendritic targeting sequence and a motif that binds to the putative tag. Glossary We can imagine that learning would leave a trail of silent LTP devices at all Contextual fear conditioning Morris water maze A hippocampus-dependent form of Pavlovian A spatial learning and memory task in which a rodent potentiated synapses (FIG. 4e), much like conditioning in which a rodent comes to associate a learns the position of an escape platform placed beneath Hänsel and Gretel left a trail of pebbles to context defined by polymodal sensory cues with an the surface of a pool of opaque water using a set of distal mark their path through the enchanted electrical footshock. visual cues. forest in the famous story by the brothers Cued fear conditioning Nictitating membrane/eyeblink conditioning Grimm. Subsequent loss of this memory, A hippocampus-independent form of Pavlovian A form of classical Pavlovian conditioning in which an either through erasure or forgetting, would conditioning in which a rodent comes to associate a tone animal gradually modifies the timing of an eyeblink to an then leave an animal in a quasi-naive state cue (conditional stimulus) with an electrical footshock anticipated unconditional stimulus, using a sound or light in which the memory would be re-installed (unconditional stimulus). Learning is assessed by the conditional stimulus. Rabbits are traditionally used for this after triggering the LTP device by injection animal’s behavioural freezing. task owing to the presence of a third eyelid, or nictitating membrane, which is not under conscious control. of the ligand, just as Hänsel and Gretel fol- De-depression lowed the pebbles to find their way home. The selective reversal of LTD by high-frequency Pattern completion Transgenic ‘forgetful mice’, which are unable stimulation. The phenomenon whereby a memory can be recalled by to form long-term memories (for example, presentation of only a subset of the cues that were Depotentiation available during the learning episode. There is evidence animals in which the IEGs that encode The selective and time-dependent reversal of already- that the CA3 subregion of the hippocampus is necessary 58 Zif268 (Ref. 60) or Arc have been inacti- potentiated synapses using low-frequency stimulation. for animals to achieve pattern completion. vated), would be particularly suited for such Note that depotentiation differs from LTD in that it has no experiments. affect on unpotentiated synapses and affects only recently Pattern separation potentiated synapses. The phenomenon whereby two similar contexts can be discriminated on the basis of subtle differences in the Limited synaptic states. The strategies Episodic memory constituent cues. Such pattern separation allows the recall we have described will work optimally if Event-related memory: the ‘what, where and when’ of only those memories that are relevant to one context or the number of synaptic states is limited memory system. Experiments in rodents are largely the other. There is evidence that the dentate gyrus is — ideally, each synapse would adopt one of restricted to the ‘what’ and ‘where’ elements. We define necessary for pattern separation. hippocampus-dependent tasks such as contextual fear three discrete states: basal, potentiated or conditioning and the Morris water maze as Radial arm maze 75 depressed . If this is not the case the problem requiring episodic-like memory. Usually an eight-armed maze that can be used for becomes more challenging, but it does not various memory tasks. Here we refer to it in the context become a lost cause. In the first case, knowl- EPSP‑spike potentiation of working memory in which each arm is baited with food. edge about the size of the change would be (E-S potentiation). A potentiation not of synaptic Working memory can be assessed by how often the animal transmission, as in LTP, but of the likelihood that action returns to an arm that it has already visited and emptied of available, and the STDP stimulus protocol potentials will be generated for a given synaptic input. food reward. would be adjusted accordingly. In the case This phenomenon usually occurs in tandem with LTP after of the free-standing LTP device, it is likely high-frequency stimulation. Reference memory that information about the size of the change Long-term spatial memory that involves reference to Forgetful mouse external cues, as is needed for succesful learning of the will be encoded in the synaptic tag, and the This is a type of genetically engineered mouse that can standard form of the Morris water maze task, in which design of the device will need to take this into learn but not consolidate hippocampus-dependent the location of the hidden platform is fixed for several days. account. Finally, although we have discussed memory. these approaches only in terms of LTP, they Spike-timing-dependent plasticity can be readily extended to accommodate Long-term depression (STDP). Plasticity in which pre- and postsynaptic cells are (LTD). The opposing process to LTP, whereby synaptic stimulated independently and the timing with which spikes homosynaptic or heterosynaptic LTD. transmission is weakened by low-frequency stimulation. are evoked in the two types of cell determines the direction LTD might serve as a learning mechanism in its own right or of plasticity. Conclusion might be a means of ensuring homeostatic stability by The study of the neural basis of memory preventing an increase in overall activity in potentiated Synaptic tagging networks. Both long-term memory and LTP require mRNA has been dominated over the past 30 years transcription and protein synthesis. However, plasticity by investigations into the molecular and Long-term potentiation changes are specific to activated synapses. A mechanism, cellular basis of synaptic plasticity. We (LTP). An experimental model of synaptic plasticity. In the termed synaptic tagging, must exist to capture newly argue here that a full understanding of hippocampus, high-frequency electrical stimulation of expressed plasticity related mRNAs or proteins specifically memory and the neural circuits responsible afferent-fibre pathways induces an enhancement of at activated synapses. One possible solution is the setting synaptic transmission that can last for months. of labile ‘tags’ at activated synapses that would capture for its acquisition, encoding and recall will recently synthesized proteins. not be achieved until instrumental and Memory mimicry conceptual tools have been developed to (MM) An experiment designed to test whether LTP-like Working memory study neural networks in the large. Further plasticity alone is sufficient to support memory, by Short-term memory, used here to describe the type of artificially installing a memory of an unexperienced event. memory that is needed for successful completion of a progress in analysing the neural basis of Also called the ‘Marilyn Monroe’ thought experiment, version of the Morris water maze experiment in which the memory will require an approach that because it could entail creating a false memory of a position of the hidden platform is changed daily (see also emphasizes the importance of the network meeting with her. radial arm maze).

P e r s pec t i v e s

synaptic plasticity is necessary for stor- of associative learning in behaving mice. J. Neurosci. 43. Hoffman, D. A., Sprengel, R. & Sakmann, B. Molecular 26, 1077–1087 (2006). dissection of hippocampal theta-burst pairing ing memories. Moreover, it is becoming 17. Whitlock, J. R., Heynen, A. J., Shuler, M. G. & Bear, potentiation. Proc. Natl Acad. Sci. USA 99, possible to envisage techniques that will M. F. Learning induces long-term potentiation in the 7740–7745 (2002). hippocampus. Science 313, 1093–1097 (2006). 44. O’Keefe, J. Hippocampus, theta, and spatial memory. permit the re-installation, at the network 18. Rumpel, S., LeDoux, J., Zador, A. & Malinow, R. Curr. Opin. Neurobiol. 3, 917–924 (1993). level, of silenced or lost memories; such Postsynaptic receptor trafficking underlying a form of 45. Doyère, V. et al. Potentiation or depression of synaptic associative learning. Science 308, 83–88 (2005). efficacy in the dentate gyrus is determined by the experiments, if successful, would establish 19. Dudek, S. M. & Bear, M. F. Homosynaptic long-term relationship between the conditioned and that synaptic plasticity is also a sufficient depression and effects of N‑Methyl‑d-aspartate unconditioned stimulus in a classical conditioning receptor blockade. Proc. Natl Acad. Sci. USA 89, paradigm in rats. Behav. Brain Res. 70, 15–29 mechanism for storing memories. 4363–4367 (1992). (1995). 20. Abraham, W. C., Bliss, T. V. & Goddard, G. V. 46. Ito, M. & Kano, M. Long-lasting depression of parallel Guilherme Neves and Tim V. P. Bliss are at the Division Heterosynaptic changes accompany long-term but not fiber-Purkinje cell transmission induced by conjunctive of Neurophysiology, Medical Research Council short-term potentiation of the perforant path in the stimulation of parallel fibers and climbing fibers in the National Institute for Medical Research, Mill Hill, anaesthetized rat. J. Physiol. 363, 335–349 (1985). cerebellar cortex. Neurosci. Lett. 33, 253–258 London, NW7 1AA, UK. 21. Andersen, P., Sundberg, S. H., Sveen, O., Swann, J. W. (1982). & Wigström, H. Possible mechanisms for long-lasting 47. Marr, D. A theory of cerebellar cortex. J. Physiol. 202, Sam F. Cooke is at The Picower Institute for Learning potentiation of synaptic transmission in hippocampal 437–470 (1969). slices from guinea-pigs. J. Physiol. 302, 463–482 48. McCormick, D. A. & Thompson, R. F. Cerebellum: and Memory, Massachusetts Institute of Technology, (1980). essential involvement in the classically conditioned 77 Massachusetts Avenue, 46‑3301, Cambridge, 22. Dan, Y. & Poo, M. M. Spike timing-dependent eyelid response. Science 223, 296–299 (1984). Massachusetts 02139, USA. plasticity of neural circuits. Neuron 44, 23–30 49. Attwell, P. J., Cooke, S. F. & Yeo, C. H. Cerebellar (2004). function in consolidation of a motor memory. Neuron Correspondence to T.V.P.B. 23. Barrionuevo, G., Shottler, F. & Lynch, G. The effects of 34, 1011–1020 (2002). e-mail: [email protected] repetitive low-frequency stimulation on control and 50. Andersson, G. & Oscarsson, O. Climbing fiber “potentiated” synaptic responses in the hippocampus. microzones in cerebellar vermis and their projection to doi:10.1038/nrn2303 Life Sci. 27, 2385–2391 (1980). different groups of cells in the lateral vestibular 1. Scoville, W. B. & Milner, B. Loss of recent memory 24. Staubli, U. & Lynch, G. Stable depression of nucleus. Exp. Brain Res. 32, 565–579 (1978). after bilateral hippocampal lesions. J. Neurol. potentiated synaptic responses in the hippocampus 51. Garwicz, M. & Ekerot, C. F. Topographical organization Neurosurg. Psychiatry 20, 11–21 (1957). with 1–5Hz stimulation. Brain Res. 513, 113–118 of the cerebellar cortical projection to nucleus 2. Tsien, J. Z., Huerta, P. T. & Tonegawa, S. The essential (1990). interpositus anterior in the cat. J. Physiol. 474, role of hippocampal CA1 NMDA receptor-dependent 25. Montgomery, J. M. & Madison, D. V. State-dependent 245–260 (1994). synaptic plasticity in spatial memory. Cell 87, heterogeneity in synaptic depression between 52. Mauk, M. D., Steinmetz, J. E. & Thompson, R. F. 1327–1338 (1996). pyramidal cell pairs. Neuron 33, 765–777 (2002). Classical conditioning using stimulation of the inferior 3. Martin, S. J., de Hoz, L. & Morris, R. G. Retrograde 26. Dudek, S. M. & Bear, M. F. Bidirectional long-term olive as the unconditioned stimulus. Proc. Natl Acad. amnesia: neither partial nor complete hippocampal modification of synaptic effectiveness in the adult and Sci. USA 83, 5349–5353 (1986). lesions in rats result in preferential sparing of remote immature hippocampus. J. Neurosci. 13, 2910–2918 53. Nakazawa, K. et al. Requirement for hippocampal CA3 spatial memory, even after reminding. (1993). NMDA receptors in associative memory recall. Science Neuropsychologia 43, 609–624 (2005). 27. Skrede, K. K. R. & Westgaard, R. H. The transverse 297, 211–218 (2002). 4. Morris, R., Anderson, E., Lynch, G. S. & Baudry, M. hippocampal slice: a well-defined cortical structure 54. McHugh, T. J. et al. Dentate gyrus NMDA receptors Selective impairment of learning and blockade of long- maintained in vitro. Brain Res. 35, 589–593 (1971). mediate rapid pattern separation in the hippocampal term potentiation by an N‑methyl‑d-aspartate receptor 28. Banker, G. A. & Cowan, W. M. Rat hippocampal network. Science 317, 94–99 (2007). antagonist, ap5. Nature 319, 774–776 (1986). neurons in dispersed cell culture. Brain Res. 126, 55. Niewoehner, B. et al. Impaired spatial working 5. Pastalkova, E. et al. Storage of spatial information by 397–342 (1977). memory but spared spatial reference memory the maintenance mechanism of LTP. Science 313, 29. Gahwiler, B. H. Organotypic monolayer cultures of following functional loss of NMDA receptors in the 1141–1144 (2006). nervous tissue. J. Neurosci. Methods 4, 329–342 dentate gyrus. Eur. J. Neurosci. 25, 837–846 6. Berger, T. W., Rinaldi, P. C., Weisz, D. J. & Thompson, (1981). (2007). R. F. Single-unit analysis of different hippocampal cell 30. Malenka, R. C. & Bear, M. F. LTP and LTD: an 56. Marr, D. Simple memory: a theory for archicortex. types during classical conditioning of rabbit nictitating embarrassment of riches. Neuron 44, 5–21 (2004). Philos. Trans. R. Soc. Lond. B Biol. Sci. 262, 23–81 membrane response. J. Neurophysiol. 50, 31. Giese, K. P., Fedorov, N. B., Filipkowski, R. K. & Silva, (1971). 1197–1219 (1983). A. J. Autophosphorylation at Thr286 of the α calcium- 57. Guzowski, J. F. et al. Mapping behaviorally relevant 7. Vazdarjanova, A. & Guzowski, J. F. Differences in calmodulin kinase II in LTP and learning. Science 279, neural circuits with immediate-early gene expression. hippocampal neuronal population responses to 870–873 (1998). Curr. Opin. Neurobiol. 15, 599–606 (2005). modifications of an environmental context: evidence 32. Hebb, D. O. The Organization of Behavior (Wiley, New 58. Plath, N. et al. Arc/Arg3.1 is essential for the for distinct, yet complementary, functions of CA3 and York, 1949). consolidation of synaptic plasticity and memories. CA1 ensembles. J. Neurosci. 24, 6489–6496 (2004). 33. O’Keefe, J. & Dostrovsky, J. The hippocampus as a Neuron 52, 437–444 (2006). 8. Guzowksi, J. F., Setlow, B., Wagner, E. K. & McGaugh, spatial map. Preliminary evidence from unit activity in 59. Wang, K. H. et al. In vivo two-photon imaging reveals J. L. Experience-dependent gene expression in the rat the freely-moving rat. Brain Res. 34, 171–175 (1971). a role of Arc in enhancing orientation specificity in hippocampus after spatial learning: a comparison of 34. Wilson, M. A. & McNaughton, B. L. Dynamics of the visual cortex. Cell 126, 389–402 (2006). the immediate-early genes Arc, c‑fos, and zif‑268. hippocampal ensemble code for space. Science 261, 60. Jones, M. W. et al. A requirement for the immediate J. Neurosci. 21, 5089–5098 (2001). 1055–1058 (1993). early gene Zif268 in the expression of late LTP and 9. Gabrieli, J. D., Brewer, J. B., Desmond, J. E. & 35. Skaggs, W. E. & McNaughton, B. L. Replay of neuronal long-term memories. Nature Neurosci. 4, 289–296 Glover, G. H. Separate neural bases of two firing sequences in rat hippocampus during sleep (2001). fundamental memory processes in the human medial following spatial experience. Science 271, 61. Tagawa, Y., Kanold, P. O., Majdan, M. & Shatz, C. J. temporal lobe. Science 276, 264–266 (1997). 1870–1873 (1996). Multiple periods of functional ocular dominance 10. Maguire, E. A. Neuroimaging, memory and the human 36. Martin, S. J., Grimwood, P. D. & Morris, R. G. Synaptic plasticity in mouse visual cortex. Nature Neurosci. 8, hippocampus. Rev. Neurol. (Paris) 157, 791–794 plasticity and memory: an evaluation of the hypothesis. 380–388 (2005). (2001). Annu. Rev. Neurosci. 23, 649–711 (2000). 62. Reijmers, L. G., Perkins, B. L., Matsuo, N. & Mayford, M. 11. Henke, K., Buck, A., Weber, B. & Wieser, H. G. Human 37. Shimizu, E., Tang, Y. P., Rampon, C. & Tsien, J. Z. Localization of a stable neural correlate of associative hippocampus establishes associations in memory. NMDA receptor-dependent synaptic reinforcement as memory. Science 317, 1230–1233 (2007). Hippocampus 7, 249–256 (1997). a crucial process for memory consolidation. Science 63. Wilson, Y. et al. Visualization of functionally activated 12. Andersen, P., Bliss, T. V., Lomo, T., Olsen, L. I. & 290, 1170–1174 (2000). circuitry in the brain. Proc. Natl Acad. Sci. USA 99, Skrede, K. K. Lamellar organization of hippocampal 38. Jester, J. M., Campbell, L. W. & Sejnowski, T. J. 3252–3257 (2002). excitatory pathways. Acta Physiol. Scand. 76, 4A–5A Associative EPSP-spike potentiation induced by 64. Barth, A. L., Gerkin, R. C. & Dean, K. L. Alteration of (1969). pairing orthodromic and antidromic stimulation in rat neuronal firing properties after in vivo experience in a 13. Bliss, T. V. & Lømo, T. Long-lasting potentiation of hippocampal slices. J. Physiol. 484, 589–705 (1995). FosGFP transgenic mouse. J. Neurosci. 24, synaptic transmission in the dentate area of the 39. Herron, C. E., Lester, R. A., Coan, E. J. & Collingridge, 6466–6475 (2004). anaesthetized rabbit following stimulation of the G. L. Frequency-dependent involvement of NMDA 65. Slimko, E., McKinney, S., Anderson, D. J., Davidson, N. perforant path. J. Physiol. 232, 331–356 (1973). receptors in the hippocampus: a novel synaptic & Lester, H. A. Selective electrical silencing of 14. Bliss, T. V. & Gardner-Medwin, A. R. Long-lasting mechanism. Nature 322, 265–268 (1986). mammalian neurons in vitro by the use of invertebrate potentiation of synaptic transmission in the dentate 40. Bannerman, D. M., Good, M. A., Butcher, S. P., ligand-gated chloride channels. J. Neurosci. 22, area of the unanaesthetized rabbit following Ramsay, M. & Morris, R. G. Distinct components of 7373–7379 (2002). stimulation of the perforant path. J. Physiol. 232, spatial learning revealed by prior training and NMDA 66. Lerchner, W. et al. Reversible silencing of neuronal 357–374 (1973). receptor blockade. Nature 378, 182–186 (1995). excitability in behaving mice by a genetically targeted, 15. Bliss, T. V. P., Collingridge, G. L. & Morris, R. G. M. in 41. King, C., Henze, D. A., Leinekugel, X. & Buzsaki, G. ivermectin-gated Cl– channel. Neuron 54, 35–49 The Hippocampus Book (eds Andersen, P., Morris, Hebbian modification of a hippocampal population (2007). R. G. M., Amaral, D. G., Bliss, T. V. P. & O’Keefe, J.) pattern in the rat. J. Physiol. 521, 159–167 (1999). 67. Lechner, H. A., Lein, E. S. & Callaway, E. M. A genetic 343–474 (Oxford Univ. Press, New York, 2007). 42. Zamanillo, D. et al. Importance of AMPA receptors for method for selective and quickly reversible silencing of 16. Gruart, A., Munoz, M. D. & Delgado-Garcia, J. M. hippocampal synaptic plasticity but not for spatial mammalian neurons. J. Neurosci. 22, 5287–5290 Involvement of the CA3-CA1 synapse in the acquisition learning. Science 284, 1805–1811 (1999). (2002).

P e r s pec t i v e s

68. Gosgnach, S. et al. V1 spinal neurons regulate the 80. Lever, C., Wills, T., Cacucci, F., Burgess, N. & O’Keefe, J. Bliss, T. V. P. & O’Keefe, J.) 37–114 (Oxford Univ. speed of vertebrate locomotor outputs. Nature 440, Long-term plasticity in hippocampal place-cell Press, New York, 2007). 215–219 (2006). representation of environmental geometry. Nature 91. Bradshaw, K. D., Emptage, N. J. & Bliss, T. V. A role for 69. Karpova, A. Y., Tervo, D. G., Gray, N. W. & Svoboda, K. 416, 90–94 (2002). dendritic protein synthesis in hippocampal late LTP. Rapid and reversible chemical inactivation of synaptic 81. Fyhn, M., Hafting, T., Treves, A., Moser, M. B. & Eur. J. Neurosci. 18, 3150–3152 (2003). transmission in genetically targeted neurons. Neuron Moser, E. I. Hippocampal remapping and grid 92. Ramirez-Amaya, V. et al. Spatial exploration-induced 48, 727–735 (2005). realignment in entorhinal cortex. Nature 446, arc mRNA and protein expression: evidence for 70. Wulff, P. et al. From synapse to behavior: rapid 190–194 (2007). selective, network-specific reactivation. J. Neurosci. modulation of defined neuronal types with engineered 82. Moita, M. A., Rosis, S., Zhou, Y., LeDoux, J. E. & Blair, 25, 1761–1768 (2005).

GABAA receptors. Nature Neurosci. 10, 923–929 H. T. Putting fear in its place: remapping of 93. Bourtchuladze, R. et al. Deficient long-term memory in (2007). hippocampal place cells during fear conditioning. mice with a targeted mutation of the cAMP-responsive 71. Zhang, F. et al. Multimodal fast optical interrogation J. Neurosci. 24, 7015–7023 (2004). element-binding protein. Cell 79, 59–68 (1994). of neural circuitry. Nature 446, 633–639 (2007). 83. McHugh, T. J., Blum, K. I., Tsien, J. Z., Tonegawa, S. & 94. Steward, O., Wallace, C. S., Lyford, G. L. & Worley, P. F. 72. Zhang, F., Aravanis, A. M., Adamantidis, A., de Lecea, L. Wilson, M. A. Impaired hippocampal representation of Synaptic activation causes the mRNA for the IEG Arc & Deisseroth, K. Circuit-breakers: optical technologies space in CA1-specific NMDAR1 knockout mice. to localize selectively near activated postsynaptic sites for probing neural signals and systems. Nature Rev. Cell 87, 1339–1349 (1996). on dendrites. Neuron 21, 741–751 (1998). Neurosci. 8, 577–581 (2007). 84. Kentros, C. et al. Abolition of long-term stability of 95. Moga, D. E. et al. Activity-regulated cytoskeletal- 73. Adamantidis, A. R., Zhang, F., Aravanis, A. M., new hippocampal place cell maps by NMDA receptor associated protein is localized to recently activated Deisseroth, K. & de Lecea, L. Neural substrates of blockade. Science 280, 2121–2126 (1998). excitatory synapses. Neuroscience 125, 7–11 (2004). awakening probed with optogenetic control of 85. Nakazawa, K. et al. Hippocampal CA3 NMDA hypocretin neurons. Nature 450, 420–424 (2007). receptors are crucial for memory acquisition of one- Acknowledgements 74. Frey, U. & Morris, R. G. Synaptic tagging and long- time experience. Neuron 38, 305–315 (2003). We would like to thank P. Skehel for his contribution to the term potentiation. Nature 385, 533–536 (1997). 86. Teyler, T. J. & Rudy, J. W. The hippocampal indexing ideas that have been developed here, and M. Bear, R. Morris, 75. Montgomery, J. M. & Madison, D. V. Discrete synaptic theory and episodic memory: updating the index. J. O’Keefe and J. Young for valuable discussions. states define a major mechanism of synapse plasticity. Hippocampus 17, 1158–1169 (2007). Trends Neurosci. 27, 744–750 (2004). 87. Squire, L. R. & Bayley, P. J. The neuroscience of 76. Deadwyler, S. A. & Hampson, R. E. Anatomic model of remote memory. Curr. Opin. Neurobiol. 17, 185–196 hippocampal encoding of spatial information. (2007). DATABASES Hippocampus 9, 397–412 (1999). 88. Nadel, L., Wincour, G., Ryan, L. & Moscovitch, M. Entrez Gene: http://www.ncbi.nlm.nih.gov/entrez/query. 77. Redish, A. D. et al. Independence of firing correlates Systems consolidation and hippocampus: two views. fcgi?db=gene of anatomically proximate hippocampal pyramidal Debates Neurosci. 6 Mar 2007 (doi:10.1007/ Arc/Arg3.1 | AlstR | c‑Fos | ChR2 | GluR1 | Homer | NpHR | NR1 | cells. J. Neurosci. 21, RC134 (2001). s11559‑007‑9003–9). Zif268 78. Moser, M. B. & Moser, E. I. Distributed encoding and 89. Morris, R. G. M. in The Hippocampus Book (eds retrieval of spatial memory in the hippocampus. Andersen, P., Morris, R. G. M., Amaral, D. G., Bliss, FURTHER INFORMATION J. Neurosci. 18, 7535–7542 (1998). T. V. P. & O’Keefe, J.) 581–714 (Oxford Univ. Press, Tim V. P. Bliss’s homepage: 79. Bostock, E., Muller, R. U. & Kubie, J. L. Experience- New York, 2007). www.nimr.mrc.ac.uk/neurophysiol/bliss dependent modifications of hippocampal place cell 90. Amaral, D. G. & Lavenex, P. in The Hippocampus Book All links are active in the online pdf firing. Hippocampus 1, 193–205 (1991). (eds Andersen, P., Morris, R. G. M., Amaral, D. G.,